Automated planning is a subdiscipline of artificial intelligence that aims at generating plans, or sequences of actions, that lead to some desired outcome. In other words, given a goal state, an initial state, and some possible actions, an automated planner outputs a plan that, when executed, will reach the goal state from the initial state.
Automated planners are used heavily in robotics, logistics, scheduling, and other areas. In some areas planning may be offline and for analysis purposes; in other areas, it may need to be real-time and for control of some real-world process or entity (think: controlling robots on factory floors, etc). Often the generated plans must be optimized for some criteria (overall cost, time, etc).
There are many formalisms that form the basis of different kinds of automated planning systems; our automated planner, HTN-DL Planner, uses the Hierarchical Task Network (HTN) planning formalism for a variety of reasons, including the fact that HTN planning generally maps well to the way people think about building plans of actions (i.e., hierarchically: you don’t worry about all of the details of returning your rental car when trying to pick the days you will travel for a business trip; that level of detail is pushed down into the leaves of the plan from a high-level, and so on).
Our HTN-DL Planner is also integrated with Pellet and uses OWL-S to describe plans, so it’s nicely integrated not only with Semantic Web but also with web service technologies. It uses Pellet and OWL to represent the state of the world and inference to make queries and data more agile.
Automated planning is like automated reasoning in one sense: they’re both artificial intelligence technologies that, within limited domains, have shown impressive utility.
They are different in another sense: automated reasoning is generally accepted as a core Semantic Web technology; and automated planning isn’t. We think that difference is unfortunate and reversible, which suggests a wider engagement and publicity effort about automated planning. But that’s effort for another day; first, let’s see if we can inject the idea into the mass consciousness about what automated planning is and what it’s good for.
